Refine model of qwen.

qwen/modeling_qwen.py

@@ -204,46 +204,20 @@ class QwenRunner:
         pad_token_id = qwen.config.pad_token_id
 
         unfinished_sequences = torch.ones(input_ids.shape[0], dtype=torch.long, device=input_ids.device)
-        this_peer_finished = False
         while True:
             outputs = self.forwardQWen(input_ids)
             next_token_scores = outputs[:, -1, :]
 
-            # repetition_penalty
+            next_token_scores = self.repetition_penalty(input_ids, next_token_scores)
-            penalty = qwen.config.repetition_penalty
+            next_token_scores = self.top_p(next_token_scores)
-            score = torch.gather(next_token_scores, 1, input_ids)
+            next_tokens = self.sample(next_token_scores)
-            # if score < 0 then repetition penalty has to be multiplied to reduce the token probabilities
-            score = torch.where(score < 0, score * penalty, score / penalty)
-            next_token_scores = next_token_scores.scatter_(1, input_ids, score)
-
-            # top_p
-            top_p = qwen.config.top_p
-            filter_value = -float("Inf")
-            min_tokens_to_keep = 1
-            sorted_logits, sorted_indices = torch.sort(next_token_scores, descending=False)
-            cumulative_probs = sorted_logits.softmax(dim=-1).cumsum(dim=-1)
-            # Remove tokens with cumulative top_p above the threshold (token with 0 are kept)
-            sorted_indices_to_remove = cumulative_probs <= (1 - top_p)
-            # Keep at least min_tokens_to_keep
-            sorted_indices_to_remove[..., -min_tokens_to_keep:] = 0
-            # scatter sorted tensors to original indexing
-            indices_to_remove = sorted_indices_to_remove.scatter(1, sorted_indices, sorted_indices_to_remove)
-            next_token_scores = next_token_scores.masked_fill(indices_to_remove, filter_value)
-
-            # sample
-            probs = nn.functional.softmax(next_token_scores, dim=-1)
-            next_tokens = torch.multinomial(probs, num_samples=1).squeeze(1)
 
             next_tokens = next_tokens * unfinished_sequences + pad_token_id * (1 - unfinished_sequences)
             input_ids = torch.cat([input_ids, next_tokens[:, None]], dim=-1)
             unfinished_sequences = unfinished_sequences.mul(
                 next_tokens.tile(eos_token_id_tensor.shape[0], 1).ne(eos_token_id_tensor.unsqueeze(1)).prod(dim=0)
             )
-
             if unfinished_sequences.max() == 0:
-                this_peer_finished = True
-
-            if this_peer_finished:
                 break
 
         decoded, response, end_reason = decode_tokens(
@@ -254,7 +228,7 @@ class QwenRunner:
             errors="replace",
         )
         history.append((query, response))
-        return response, history, decoded
+        return input_ids[0].cpu().tolist(), history, decoded
 
     def _rotate_half(self, x):
         x = rearrange(x, "... (j d) -> ... j d", j=2)
@@ -379,3 +353,31 @@ class QwenRunner:
         # loss.backward()
 
         return lm_logits
+
+    def repetition_penalty(self, input_ids, next_token_scores):
+        penalty = self.qwen.config.repetition_penalty
+        score = torch.gather(next_token_scores, 1, input_ids)
+        # if score < 0 then repetition penalty has to be multiplied to reduce the token probabilities
+        score = torch.where(score < 0, score * penalty, score / penalty)
+        next_token_scores = next_token_scores.scatter_(1, input_ids, score)
+        return next_token_scores
+
+    def top_p(self, next_token_scores):
+        top_p = self.qwen.config.top_p
+        filter_value = -float("Inf")
+        min_tokens_to_keep = 1
+        sorted_logits, sorted_indices = torch.sort(next_token_scores, descending=False)
+        cumulative_probs = sorted_logits.softmax(dim=-1).cumsum(dim=-1)
+        # Remove tokens with cumulative top_p above the threshold (token with 0 are kept)
+        sorted_indices_to_remove = cumulative_probs <= (1 - top_p)
+        # Keep at least min_tokens_to_keep
+        sorted_indices_to_remove[..., -min_tokens_to_keep:] = 0
+        # scatter sorted tensors to original indexing
+        indices_to_remove = sorted_indices_to_remove.scatter(1, sorted_indices, sorted_indices_to_remove)
+        next_token_scores = next_token_scores.masked_fill(indices_to_remove, filter_value)
+        return next_token_scores
+
+    def sample(self, next_token_scores):
+        probs = nn.functional.softmax(next_token_scores, dim=-1)
+        next_tokens = torch.multinomial(probs, num_samples=1).squeeze(1)
+        return next_tokens

qwen/research_attention.py

@@ -70,34 +70,48 @@ def Dump_lm_head_weight(model):
 # Dump_lm_head_weight(model)
 
 
+def DumpQK(query, key, causal_mask, index):
+    scale_factor = 1 / math.sqrt(query.size(-1))
+    attn_weight = query @ key.transpose(-2, -1) * scale_factor
+    attn_weight = torch.softmax(attn_weight, dim=-1)
+    size = query.shape[2]
+    attn_mask = torch.ones(causal_mask.shape, dtype=query.dtype, device=query.device)
+    attn_mask.masked_fill_(causal_mask.logical_not(), float(0))
+    qk = attn_weight * attn_mask
+    qk = qk[0]
+    prePath = "./temp/" + "q@k_seq_" + str(size) + "_layer_" + str(index) + ".png"
+    show.DumpTensorToImage(qk, prePath, GridValue=255)
+
+
 class ResearchRunner(QwenRunner):
+    def __init__(self, model):
+        super().__init__(model)
+
     def attention(self, attention, query, key, value, causal_mask):
         query = query.permute(0, 2, 1, 3)
         key = key.permute(0, 2, 1, 3)
         value = value.permute(0, 2, 1, 3)
 
-        scale_factor = 1 / math.sqrt(query.size(-1))
+        DumpQK(query, key, causal_mask, attention.index)
-        attn_weight = query @ key.transpose(-2, -1) * scale_factor
-        attn_weight = torch.softmax(attn_weight, dim=-1)
-        size = query.shape[2]
-        attn_mask = torch.ones(causal_mask.shape, dtype=query.dtype, device=query.device)
-        attn_mask.masked_fill_(causal_mask.logical_not(), float(0))
-        qk = attn_weight * attn_mask
-        qk = qk[0]
-        prePath = "./temp/" + "q@k_seq_" + str(size) + "_layer_" + str(attention.index) + ".png"
-        show.DumpTensorToImage(qk, prePath, GridValue=255)
 
         attn_output = F.scaled_dot_product_attention(query, key, value, attn_mask=causal_mask).transpose(1, 2)
         context_layer = attention._merge_heads(attn_output, attention.num_heads, attention.head_dim)
         attn_output = attention.c_proj(context_layer)
         return attn_output
 
-
 runner = ResearchRunner(model)
 
 # 第一轮对话
-response, history, decode_tokens = runner.Chat(tokenizer, "东南亚国家日本的首都是什么市", "日本的首都是")
+output_ids, history, decoded = runner.Chat(tokenizer, "东南亚国家日本的首都是什么市", "日本的首都是")
-print(decode_tokens)
+print(decoded)
+
+tokens = []
+for i, token in enumerate(output_ids):
+    de = tokenizer.decode([token])
+    de = str(i).zfill(3) + " : " + repr(de)
+    tokens.append(de)
+    print(de)
+
 # <|im_start|>system
 # You are a helpful assistant.<|im_end|>
 # <|im_start|>user
@@ -106,5 +120,7 @@ print(decode_tokens)
 # 日本的首都东京。<|im_end|>
 # <|endoftext|>
 
-if decode_tokens.split("\n")[-2] != """日本的首都东京。<|im_end|>""":
+show.DumpListToFile(tokens, "./temp/token_decode_list.txt")
+
+if decoded.split("\n")[-2] != """日本的首都东京。<|im_end|>""":
     raise ()